scipy does not exit if there are build failures, but spkg-install looks OK

[sagetrac-drkirkby]

Building Sage 4.5 on a Sun Blade 2000, with dual UltraSPARC III+ processors in 64-bit mode, the build process produces some obvious error messages. These are not warnings, but errors.

gcc _configtest.o -L/export/home/drkirkby/64/sage-4.5/local/lib -lf77blas -lcblas -latlas -o _configtest
ld: fatal: file _configtest.o: wrong ELF class: ELFCLASS64
ld: fatal: File processing errors. No output written to _configtest
collect2: ld returned 1 exit status
ld: fatal: file _configtest.o: wrong ELF class: ELFCLASS64
ld: fatal: File processing errors. No output written to _configtest
collect2: ld returned 1 exit status
failure.
removing: _configtest.c _configtest.o
Status: 255
Output:

wrong ELF class: messages mean an attempt was made to link a mixture of 32-bit and 64-bit object files.

But the build process still goes on to report that scipy has installed OK.

real    22m34.927s
user    20m23.356s
sys     1m5.603s
Successfully installed scipy-0.7.p5

What is odd, is that spkg-install looks to be OK to me.

python setup.py build
if [ $? -ne 0 ]; then
    echo "Error building scipy."
    exit 1
fi

# Intall
python setup.py install
if [ $? -ne 0 ]; then
    echo "Error installing scipy."
    exit 1
fi

The problem is not like the cephes package, or several others, where the return code of make is not checked.

At this point, I've not no idea if this is an upstream bug, or a Sage bug.

Anyone got any ideas?

Dave

Upstream: None of the above - read trac for reasoning.

CC: @nexttime @qed777 @sagetrac-mvngu @jhpalmieri @dimpase

Component: build

Reviewer: Dima Pasechnik

Issue created by migration from https://trac.sagemath.org/ticket/9519

[qed777]

comment:2

Could you give a link to the full log for scipy?

[nexttime]

comment:3

Well, _configtest sounds rather harmless...

Are you sure the finally produced(?) scipy is really broken?

[sagetrac-drkirkby]

comment:4

Replying to @qed777:

Could you give a link to the full log for scipy?

Sure. Will attach to the ticket. Since its fairly large, I have compressed it.

I think there is a problem with the build of ATLAS, as that shows a wrong ELF class: message. Since scipy links to ATLAS, that is probably the problem. But I don't think that is an excuse for the scipy build to continue, though Leif thinks it's harmless.

Dave

[sagetrac-drkirkby]

Attachment: scipy-0.7.p5.log.gz

Compressed version of sage-4.5/spkg/logs/scipy-0.7.p5.log

[qed777]

comment:5

If enough parts of Sage run, you could try running the test suite, e.g.,

$ ./sage -sh
sage subshell$ easy_install nose
sage subshell$ exit
$ ./sage -python -c "import scipy; scipy.test()"

Here's the output on sage.math with 4.5.

I suppose we could optionally (if the user/developer desires) install nose automatically.

[sagetrac-drkirkby]

comment:6

Replying to @nexttime:

Well, _configtest sounds rather harmless...

Are you sure the finally produced(?) scipy is really broken?

I'm not certain. I thought ATLAS was broken, but I think I have remade the broken library. How can I test just scipy? The 64-bit SPARC port is not 100% ok. I can't run the doctests, as it segfaults, but Sage does semi-work. I can do computations with it. So sage -t foobar might work

Dave

[sagetrac-drkirkby]

comment:7

Replying to @qed777:

If enough parts of Sage run, you could try running the test suite, e.g.,

$ ./sage -sh
sage subshell$ easy_install nose
sage subshell$ exit
$ ./sage -python -c "import scipy; scipy.test()"

Here's the output on sage.math with 4.5.

I suppose we could optionally (if the user/developer desires) install nose automatically.

I've never come across 'nose' - my python skills are next to zero. I will attach a log. As you can see, it finally fails with a core dump, but perhaps has some useful information before it dumps core.

BTW, I've removed the static atlas libs, leaving only the shared ones. That might be a cause of a problem. I can't understand the need for both. If yyou look in the ATLAS package, you can see about 3 dynamic libraries will be missing. Two are not built, and one is deleted. I built them all and deleted no dynamic ones - only the static.

Dave

[sagetrac-drkirkby]

Attachment: scipy.test.log

Output from: ./sage -python -c "import scipy; scipy.test()" > scipy.test.log 2>&1

[mkoeppe]

comment:12

outdated, should be closed

[dimpase]

Reviewer: Dima Pasechnik

[dimpase]

comment:13

ok